Statistical & Math
Forest Plot Meta-Analysis: Pooled Effect, 95% CI & Inverse-Variance Weights
Build clinical-trial Forest Plot summary rows by exponentiating log-odds/log-risk effects EXP(theta +/- 1.96 * SE), computing fixed-effect inverse-variance weights w_i = 1 / SE_i^2, and pooling theta_pooled = SUM(w_i * theta_i) / SUM(w_i).
Why Forest Plot error bars MUST be calculated on the log scale before exponentiating
For an illustrative fixed-effect inverse-variance calculation, take each study estimate theta as a log odds ratio and each positive standard error SE on that same log scale. Approximate Wald 95% intervals are EXP(theta ± 1.96 × SE); weight each estimate by 1/SE² and exponentiate the weighted mean. For the three supplied example studies, the pooled odds ratio rounds to 0.70 and the upper 95% Wald bound to 0.88. These are model-based approximations, not exact intervals or a treatment recommendation. Fixed-effect pooling assumes a common effect and independent study estimates; it does not address heterogeneity, study quality, publication bias, or correlated estimates. Odds reductions are not the same as risk reductions.
To build exact Forest Plot whiskers and fixed-effect inverse-variance pooling across K studies: (1) each study’s 95% CI spans [EXP(theta_i - 1.96 * SE_i), EXP(theta_i + 1.96 * SE_i)], (2) each study receives precision weight w_i = 1 / (SE_i ^ 2), and (3) the pooled summary diamond sits at OR_pooled = EXP(SUM(w_i * theta_i) / SUM(w_i)) with pooled standard error SE_pooled = 1 / SQRT(SUM(w_i)). Across three trials with log-ORs [-0.35, -0.52, -0.28] and SEs [0.20, 0.25, 0.16] (weights [25.00, 16.00, 39.0625]), pooled log-OR = -0.3498 -> OR_pooled = 0.70 (a statistically significant 30% odds reduction, 95% CI [0.57, 0.88] < 1.00).
| Clinical Trial | Log-OR (theta_i) & SE_i | Precision Weight wi = 1 / SEi^2 | Study OR & 95% Forest Plot Whiskers |
|---|---|---|---|
| Trial Alpha | -0.35 (SE = 0.20) | 25.00 (31.2% share) | OR 0.70 [0.48, 1.04] |
| Trial Beta | -0.52 (SE = 0.25) | 16.00 (20.0% share) | OR 0.59 [0.36, 0.97] |
| Trial Gamma | -0.28 (SE = 0.16) | 39.06 (48.8% share) | OR 0.76 [0.55, 1.03] -> Pooled OR = 0.70 |
Inverse-Variance Fixed-Effect Pooled Odds Ratio Across 3 Trials
Weights each study log-effect B_i by w_i = 1 / C_i^2 and exponentiates the precision-weighted mean.
=IF(OR(COUNT(B2:B4) <> 3, COUNT(C2:C4) <> 3, MIN(C2:C4) <= 0), "Missing/Invalid Study Input", ROUND(EXP(((B2 / (C2 ^ 2)) + (B3 / (C3 ^ 2)) + (B4 / (C4 ^ 2))) / ((1 / (C2 ^ 2)) + (1 / (C3 ^ 2)) + (1 / (C4 ^ 2)))), 2))Upper 95% Forest Plot Diamond Bound: EXP(theta_pooled + 1.96 / SQRT(SUM(w_i)))
Computes a rounded upper bound for this model-based Wald interval. A 95% interval excluding 1 corresponds to a two-sided 5% Wald test under its assumptions; assess the unrounded values and the appropriateness of the model.
=IF(OR(COUNT(B2:B4) <> 3, COUNT(C2:C4) <> 3, MIN(C2:C4) <= 0), "Missing/Invalid Study Input", ROUND(EXP((((B2 / (C2 ^ 2)) + (B3 / (C3 ^ 2)) + (B4 / (C4 ^ 2))) / ((1 / (C2 ^ 2)) + (1 / (C3 ^ 2)) + (1 / (C4 ^ 2)))) + (1.96 / SQRT((1 / (C2 ^ 2)) + (1 / (C3 ^ 2)) + (1 / (C4 ^ 2))))), 2))Try the example
Use the original values or change them to test your understanding.